Navigating the social world requires people to predict others' actions. This poses a significant challenge be- cause people cannot directly observe some of the best predictors of action: others' internal mental states. In- dividuals who can leverage information about these hidden causes of actions?by representing mental states?can better predict those actions and more successfully navigate the social world. How do people (i) represent the richness and complexity of others' invisible mental states, and (ii) use those representations to make social predictions? We propose that people reduce the complexity of others' minds by attending to the location of their mental states on a few key dimensions in a mental state ?map.? We have previously used rep- resentational similarity analyses (RSA) on functional neuroimaging (fMRI) data to show that people indeed represent others' mental states using a simple, low-dimensional map. The structure of this map is defined by three dimensions?rationality, social impact, and valence. Understanding how people employ this map will provide key insights into how people predict others' actions. Aim 1: This proposal seeks to develop a compre- hensive framework of mental state representations by first characterizing the structure of the map of mental states. We will refine the structure by assessing how it adapts across new social contexts and modalities. We will measure two structural features of the map?size and shape?using novel RSA methods on fMRI data. Di- mensions that have universal social functions should hold a stable shape across all contexts and modalities; dimensions that have specific, or contextualized functions should deform across context or modality. The size of the space should expand to reflect the social relevance of the target. Understanding the structure of the men- tal state map lays the foundation for understanding how people represent others' mental states. Aim 2: We next explore how people leverage this map to make social predictions. We propose that the mental state map encodes not only the location of others' current mental state, but also where in the map they will likely move to Thus, people could make social predictions We will use fMRI, large-scale experience sampling studies, and computational modeling over behavioral data to establish that people indeed spontaneously model others' mental state dynamics, and moreover, that these models make accurate social predictions. In both aims, we will test how the structure and dynamic of mental state maps predict social functioning (or dysfunction). Using an individual differences ap- next. by modeling the dynamics of others' mental states as paths through this map. proach, we will link our novel measures of structure and dynamics to participants' performance on a battery of social cognition, social behavior, and social relationship measures. Taken together, this proposal uses innova- tive techniques (e.g., novel RSA methods, Markov modeling, and experience sampling) to develop a compre- hensive theory of mental state representations and social predictions. Understanding these basic building blocks of social cognition carries promising future directions for assessing and ameliorating social deficits.